Start here — where these packs come from

How the research works Shipped

Every pack below was produced by researchteam, the same way you would produce one. The method is worth a minute before the findings.

These are not a hand-curated bibliography. Each pack is the output of myteams researchteam research "<question>" — a run that fans out across the team's specialists, gathers sources, and returns claims ranked by relevance with a confidence on each. The packs are committed to the repository as markdown, which is why they can be quoted here without being paraphrased into something more flattering.

What one run does

FrameThe question is decomposed into the domains that can actually answer it.
GatherEach specialist reads in its own domain, with the web access it was granted.
AttributeA claim without a source is not a claim. Everything carries where it came from.
ScoreConfidence per claim, on a bar appropriate to the domain.
ReportRanked claims, plus an explicit list of what is still open.

What a pack looks like

Screenshot pendingThe front page of a committed research pack: the question at the top, the run id and cost beneath it, a claims table with a confidence column, and an explicit list of open questions below that.

Every pack is markdown in the repository, which is why the confidence column and the open-questions list can be quoted here rather than paraphrased. The run whose web access was denied carries its unverified count on this same front page, in the same place.

How to read a pack

Confidence is domain-relative. A claim about a published algorithm and a claim about what a vendor shipped last week do not deserve the same bar, and the same number does not mean the same thing across them.
Open questions are content. The section listing what the run could not settle is the most useful part of a pack, and it is the first thing a summary would delete.
"Changed the product?" is answered honestly. Two of the packs below changed nothing. One of them was written after the design it supposedly informed already existed.

The one that failed is still published

A run whose web access was denied reported 119 unverified claims rather than quietly presenting recall as research. It is in the repository with that count attached. A failed research run with a stated cause is more useful than no record of it — and it is the reason the tool grant is now something the team checks rather than assumes.